Chaotic Time Series Prediction Based on Fuzzy Information Granulation and Hybrid Neural Network

To improve the prediction accuracy of chaotic time series, we propose a hybrid neural network prediction model based on fuzzy information granulation and an attention mechanism. First, we normalize the data, simplify the complexity of the data using fuzzy informational granulation (FIG), and then add the samples reconstructed from phase space into a convolution neural network (CNN) to extract spatial features. Next, we use the long short-term memory network (LSTM) to further extract the temporal features. Finally, we transmit the fusion features to the attention mechanism to extract the key features and obtain the prediction results. We select Logistic, Lorenz, and sunspot chaotic time series for experiments and compare the model with the CNN-LSTM-Att, CNN-LSTM, FIG-CNN, FIG-LSTM, CNN, and LSTM models, and support vector machine (SVM) and error back propagation (BP) neural network model. The results show that the proposed prediction model has a higher prediction accuracy and a smaller error value.